getcheapaicredits

Model vs model · live prices

GPT-6 Luna vs Devstral 2 2512

GPT-6 Luna is 4.0× cheaper than Devstral 2 2512 at a 3:1 input:output mix ($0.2 vs $0.8 per million tokens).
Per 1M tokensOpenAI: GPT-6 LunaMistral: Devstral 2 2512
Input$0.1$0.4
Output$0.5$2
Cached input$0.01$0.04
Blended 3:1$0.2$0.8
Context window1.1M262K
ReleasedSeptember 22, 2026December 9, 2025
Tools · reasoningTools · ReasoningTools · No reasoning

List prices from OpenRouter, refreshed hourly. Lower price and larger context highlighted.

What the same work costs on each

At list price, before batch or caching discounts.
WorkloadGPT-6 LunaDevstral 2 2512Difference
1,000 chat messages
1,000 input + 500 output tokens each
$0.35$1.40$1.05
1,000 RAG questions
8,000 input + 500 output tokens each
$1.05$4.20$3.15
100 coding-agent steps
30,000 input + 2,000 output tokens each, no caching
$0.40$1.60$1.20

Which one should you use?

Price is only half the decision: run both on a sample of your own prompts and compare quality, speed and failure rate. If GPT-6 Luna is good enough for a task, it saves $1.05 per thousand chat messages. A common setup is to route easy requests to the cheaper model and keep Devstral 2 2512 for the hard ones.

Both are available behind one OpenAI-compatible key through OpenRouter or a reseller, so switching is a one-line change. See where credits cost least today and how to cut the token bill.

Related comparisons

FAQ